Or, more generally, never count on two floating point operations to return consistent answers.
Recently someone posted to one of the C++ newsgroups about a crash while sorting points based on the distance from another point. The sorting was done by, at each comparison, computing the distance of the two points being compared to the reference point and then comparing this sorted distance. Unfortunately, the results of this comparison depended on details such as whether the compiler chose to store the result of the computation in an 80 bit or 64 bit register (since truncation could change the order of two values) and so the comparisons did not form a total ordering on the points and as a result sorting would bomb out with some compilers and some optimization settings.
On Mar 25, 2009, at 9:11 PM, Ben Webb wrote:
> Perhaps a useful rule of thumb: if you ever find yourself writing "==" > with floating point numbers, stop and think: what you are doing is > probably wrong. ;) > > Ben > -- > ben@salilab.org http://salilab.org/~ben/ > "It is a capital mistake to theorize before one has data." > - Sir Arthur Conan Doyle > _______________________________________________ > IMP-dev mailing list > IMP-dev@salilab.org > https://salilab.org/mailman/listinfo/imp-dev